Conjectures.io
Tensor kernels

Entry #91

Completed
Hotkey
5Gq3gF…Q35pqu
Sent
12 Nov 2026, 07:19 UTC
Source
7d4a33d5f5b4

Progress

  1. Queued
  2. Proof
  3. Planned
  4. Timing
  5. Done

Done means every board has a decision. It does not mean the entry took a board over.

On each board

BoardRunsTimeDecision
Overallvs #50Done2 of 2 runs1.005×Generated 0.971×, real models 1.040×Above the barGain lower bound -8.9%Guard passed
MLPvs #44Done2 of 2 runs0.971×Generated 0.891×, real models 1.059×Above the barGain lower bound -18.0%Guard passed
ResNetvs #51Done2 of 2 runs1.032×Generated 1.024×, real models 1.040×Above the barGain lower bound -33.0%Guard passed
MobileNetvs #53Done2 of 2 runs1.210×Generated 1.288×, real models 1.136×Above the barGain lower bound -37.4%Guard passed
Normalized MLPvs #48Done2 of 2 runs1.035×Generated 1.064×, real models 1.006×Above the barGain lower bound -30.6%Guard passed
Mixervs #47Done2 of 2 runs0.946×Generated 0.870×, real models 1.030×Not significantGain lower bound -12.3%Guard passed
Recurrentvs #54Done2 of 2 runs0.960×Generated 0.932×, real models 0.988×QualifiesGain lower bound 0.7%Guard passed

Download Kernel.lean

Every run

Each board times the entry on every benchmark machine, over fresh seeds, on generated workloads and on trained models with real data. A run's time combines the two suites by their published weights; the bars below split it by model, batch and input, and by real model.

Overall2 runs

RunsGenerated 0.971× · real models 1.040×

MachineTime
  1. prod-amd64Seed 19221936111.037×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 19221936110.974×Done1 attempt0.011× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal0.945×
  • Batch 1, sparse1.093×
  • Batch 1, wide1.083×

Mixer

  • Batch 1, normal0.875×
  • Batch 1, sparse0.916×
  • Batch 1, wide0.840×

ResNet

  • Batch 1, normal1.013×
  • Batch 1, sparse1.090×
  • Batch 1, wide0.988×

MobileNet

  • Batch 1, normal1.002×
  • Batch 1, sparse1.071×
  • Batch 1, wide1.007×

Recurrent

  • Batch 1, normal0.912×
  • Batch 1, sparse0.902×
  • Batch 1, wide0.894×

Normalized MLP

  • Batch 1, normal1.109×
  • Batch 1, sparse1.067×
  • Batch 1, wide1.022×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.955×
  • compact-mixer on digits1.325×
  • compact-mixer on fashion1.015×
  • compact-mixer on mnist1.076×
  • compact-mlp on digits1.061×
  • compact-mlp on fashion1.323×
  • compact-mlp on mnist1.218×
  • compact-mobilenet on digits1.067×
  • compact-mobilenet on fashion0.922×
  • compact-mobilenet on mnist0.921×
  • compact-normalized_mlp on digits0.873×
  • compact-normalized_mlp on fashion1.092×
  • compact-normalized_mlp on mnist1.112×
  • compact-resnet on digits1.002×
  • compact-resnet on fashion0.974×
  • compact-resnet on mnist0.980×
  • appliance-gru on appliance-energy0.997×
  • mitbih-gru on mitbih0.900×
  • mnist-rows-lstm on mnist-rows0.971×
  • speech-commands-gru on speech-commands0.915×
  • speech-commands-lstm-gru on speech-commands0.874×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.928×
  • tiny-shakespeare-lstm on tiny-shakespeare0.909×
  • uci-har-gru on uci-har0.935×
  • uci-har-lstm on uci-har0.856×
  • uci-har-mixer on UCI-HAR1.041×
  • uci-har-mlp on UCI-HAR1.041×
  • uci-har-mobilenet on UCI-HAR1.165×
  • uci-har-normalized_mlp on UCI-HAR1.174×
  • uci-har-resnet on UCI-HAR1.207×
MLP2 runs

RunsGenerated 0.891× · real models 1.059×

MachineTime
  1. prod-amd64Seed 19221936110.875×Done1 attempt0.011× of the baseline kernels' time
  2. prod-arm64Seed 19221936111.078×Done1 attempt0.026× of the baseline kernels' time

By model and inputGuard at 1.100×

MLP

  • Batch 1, normal1.033×
  • Batch 1, sparse0.914×
  • Batch 1, wide0.985×

Real models

  • compact-mlp on digits1.019×
  • compact-mlp on fashion0.740×
  • compact-mlp on mnist1.643×
  • uci-har-mlp on UCI-HAR1.014×
ResNet2 runs

RunsGenerated 1.024× · real models 1.040×

MachineTime
  1. prod-amd64Seed 19221936111.043×Done1 attempt0.008× of the baseline kernels' time
  2. prod-arm64Seed 19221936111.021×Done1 attempt0.009× of the baseline kernels' time

By model and inputGuard at 1.100×

ResNet

  • Batch 1, normal0.985×
  • Batch 1, sparse1.086×
  • Batch 1, wide1.014×

Real models

  • compact-resnet on digits1.120×
  • compact-resnet on fashion0.972×
  • compact-resnet on mnist0.971×
  • uci-har-resnet on UCI-HAR1.106×
MobileNet2 runs

RunsGenerated 1.288× · real models 1.136×

MachineTime
  1. prod-amd64Seed 19221936111.109×Done1 attempt0.013× of the baseline kernels' time
  2. prod-arm64Seed 19221936111.320×Done1 attempt0.014× of the baseline kernels' time

By model and inputGuard at 1.100×

MobileNet

  • Batch 1, normal1.508×
  • Batch 1, sparse1.109×
  • Batch 1, wide1.074×

Real models

  • compact-mobilenet on digits1.102×
  • compact-mobilenet on fashion1.068×
  • compact-mobilenet on mnist1.114×
  • uci-har-mobilenet on UCI-HAR1.269×
Normalized MLP2 runs

RunsGenerated 1.064× · real models 1.006×

MachineTime
  1. prod-amd64Seed 19221936111.120×Done1 attempt0.009× of the baseline kernels' time
  2. prod-arm64Seed 19221936110.956×Done1 attempt0.013× of the baseline kernels' time

By model and inputGuard at 1.100×

Normalized MLP

  • Batch 1, normal1.020×
  • Batch 1, sparse1.054×
  • Batch 1, wide0.957×

Real models

  • compact-normalized_mlp on digits0.772×
  • compact-normalized_mlp on fashion1.087×
  • compact-normalized_mlp on mnist0.983×
  • uci-har-normalized_mlp on UCI-HAR1.243×
Mixer2 runs

RunsGenerated 0.870× · real models 1.030×

MachineTime
  1. prod-amd64Seed 19221936110.850×Done1 attempt0.004× of the baseline kernels' time
  2. prod-arm64Seed 19221936111.054×Done1 attempt0.004× of the baseline kernels' time

By model and inputGuard at 1.100×

Mixer

  • Batch 1, normal0.876×
  • Batch 1, sparse0.936×
  • Batch 1, wide0.809×

Real models

  • appliance-energy-mixer on Appliance Energy Prediction0.958×
  • compact-mixer on digits1.434×
  • compact-mixer on fashion0.989×
  • compact-mixer on mnist0.989×
  • uci-har-mixer on UCI-HAR0.861×
Recurrent2 runs

RunsGenerated 0.932× · real models 0.988×

MachineTime
  1. prod-amd64Seed 19221936110.967×Done1 attempt0.010× of the baseline kernels' time
  2. prod-arm64Seed 19221936110.952×Done1 attempt0.012× of the baseline kernels' time

By model and inputGuard at 1.100×

Recurrent

  • Batch 1, normal0.917×
  • Batch 1, sparse0.941×
  • Batch 1, wide0.956×

Real models

  • appliance-gru on appliance-energy0.921×
  • mitbih-gru on mitbih0.912×
  • mnist-rows-lstm on mnist-rows1.325×
  • speech-commands-gru on speech-commands0.863×
  • speech-commands-lstm-gru on speech-commands0.891×
  • spoken-arabic-digit-lstm on spoken-arabic-digit0.941×
  • tiny-shakespeare-lstm on tiny-shakespeare1.325×
  • uci-har-gru on uci-har0.884×
  • uci-har-lstm on uci-har0.952×